Toward Learned Chemical Perception of Force Field Typing Rules

Comprehensive hepatotoxicity prediction: ensemble model integrating machine learning and deep learning

Background

Chemicals may lead to acute liver injuries, posing a serious threat to human health. Achieving the precise safety profile of a compound is challenging due to the complex and expensive testing procedures. In silico approaches will aid in identifying the potential risk of drug candidates in the initial stage of drug development and thus mitigating the developmental cost.

Methods

In current studies, QSAR models were developed for hepatotoxicity predictions using the ensemble strategy to integrate machine learning (ML) and deep learning (DL) algorithms using various molecular features. A large dataset of 2588 chemicals and drugs was randomly divided into training (80%) and test (20%) sets, followed by the training of individual base models using diverse machine learning or deep learning based on three different kinds of descriptors and fingerprints. Feature selection approaches were employed to proceed with model optimizations based on the model performance. Hybrid ensemble approaches were further utilized to determine the method with the best performance.

Results

The voting ensemble classifier emerged as the optimal model, achieving an excellent prediction accuracy of 80.26%, AUC of 82.84%, and recall of over 93% followed by bagging and stacking ensemble classifiers method. The model was further verified by an external test set, internal 10-fold cross-validation, and rigorous benchmark training, exhibiting much better reliability than the published models.

Conclusion

The proposed ensemble model offers a dependable assessment with a good performance for the prediction regarding the risk of chemicals and drugs to induce liver damage.

The Open Force Field Initiative: Open Software and Open Science for Molecular Modeling

Force fields are a key component of physics-based molecular modeling, describing the energies and forces in a molecular system as a function of the positions of the atoms and molecules involved. Here, we provide a review and scientific status report on the work of the Open Force Field (OpenFF) Initiative, which focuses on the science, infrastructure and data required to build the next generation of biomolecular force fields. We introduce the OpenFF Initiative and the related OpenFF Consortium, describe its approach to force field development and software, and discuss accomplishments to date as well as future plans. OpenFF releases both software and data under open and permissive licensing agreements to enable rapid application, validation, extension, and modification of its force fields and software tools. We discuss lessons learned to date in this new approach to force field development. We also highlight ways that other force field researchers can get involved, as well as some recent successes of outside researchers taking advantage of OpenFF tools and data.

Regularized by Physics: Graph Neural Network Parametrized Potentials for the Description of Intermolecular Interactions

Simulations of molecular systems using electronic structure methods are still not feasible for many systems of biological importance. As a result, empirical methods such as force fields (FF) have become an established tool for the simulation of large and complex molecular systems. The parametrization of FF is, however, time-consuming and has traditionally been based on experimental data. Recent years have therefore seen increasing efforts to automatize FF parametrization or to replace FF with machine-learning (ML) based potentials. Here, we propose an alternative strategy to parametrize FF, which makes use of ML and gradient-descent based optimization while retaining a functional form founded in physics. Using a predefined functional form is shown to enable interpretability, robustness, and efficient simulations of large systems over long time scales. To demonstrate the strength of the proposed method, a fixed-charge and a polarizable model are trained on ab initio potential-energy surfaces. Given only information about the constituting elements, the molecular topology, and reference potential energies, the models successfully learn to assign atom types and corresponding FF parameters from scratch. The resulting models and parameters are validated on a wide range of experimentally and computationally derived properties of systems including dimers, pure liquids, and molecular crystals.

Chemoinformatics and artificial intelligence colloquium: progress and challenges in developing bioactive compounds

We report the main conclusions of the first Chemoinformatics and Artificial Intelligence Colloquium, Mexico City, June 15-17, 2022. Fifteen lectures were presented during a virtual public event with speakers from industry, academia, and non-for-profit organizations. Twelve hundred and ninety students and academics from more than 60 countries. During the meeting, applications, challenges, and opportunities in drug discovery, de novo drug design, ADME-Tox (absorption, distribution, metabolism, excretion and toxicity) property predictions, organic chemistry, peptides, and antibiotic resistance were discussed. The program along with the recordings of all sessions are freely available at https://www.difacquim.com/english/events/2022-colloquium/ .

End-to-end differentiable construction of molecular mechanics force fields

Molecular mechanics (MM) potentials have long been a workhorse of computational chemistry. Leveraging accuracy and speed, these functional forms find use in a wide variety of applications in biomolecular modeling and drug discovery, from rapid virtual screening to detailed free energy calculations. Traditionally, MM potentials have relied on human-curated, inflexible, and poorly extensible discrete chemical perception rules (atom types) for applying parameters to small molecules or biopolymers, making it difficult to optimize both types and parameters to fit quantum chemical or physical property data. Here, we propose an alternative approach that uses graph neural networks to perceive chemical environments, producing continuous atom embeddings from which valence and nonbonded parameters can be predicted using invariance-preserving layers. Since all stages are built from smooth neural functions, the entire process-spanning chemical perception to parameter assignment-is modular and end-to-end differentiable with respect to model parameters, allowing new force fields to be easily constructed, extended, and applied to arbitrary molecules. We show that this approach is not only sufficiently expressive to reproduce legacy atom types, but that it can learn to accurately reproduce and extend existing molecular mechanics force fields. Trained with arbitrary loss functions, it can construct entirely new force fields self-consistently applicable to both biopolymers and small molecules directly from quantum chemical calculations, with superior fidelity than traditional atom or parameter typing schemes. When adapted to simultaneously fit partial charge models, espaloma delivers high-quality partial atomic charges orders of magnitude faster than current best-practices with low inaccuracy. When trained on the same quantum chemical small molecule dataset used to parameterize the Open Force Field ("Parsley") openff-1.2.0 small molecule force field augmented with a peptide dataset, the resulting espaloma model shows superior accuracy vis-á-vis experiments in computing relative alchemical free energy calculations for a popular benchmark. This approach is implemented in the free and open source package espaloma, available at https://github.com/choderalab/espaloma.

Comparison of the United- and All-Atom Representations of (Halo)alkanes Based on Two Condensed-Phase Force Fields Optimized against the Same Experimental Data Set

The level of accuracy that can be achieved by a force field is influenced by choices made in the interaction-function representation and in the relevant simulation parameters. These choices, referred to here as functional-form variants (FFVs), include for example the model resolution, the charge-derivation procedure, the van der Waals combination rules, the cutoff distance, and the treatment of the long-range interactions. Ideally, assessing the effect of a given FFV on the intrinsic accuracy of the force-field representation requires that only the specific FFV is changed and that this change is performed at an optimal level of parametrization, a requirement that may prove extremely challenging to achieve in practice. Here, we present a first attempt at such a comparison for one specific FFV, namely the choice of a united-atom (UA) versus an all-atom (AA) resolution in a force field for saturated acyclic (halo)alkanes. Two force-field versions (UA vs AA) are optimized in an automated way using the CombiFF approach against 961 experimental values for the pure-liquid densities ρ_liq and vaporization enthalpies ΔH_vap of 591 compounds. For the AA force field, the torsional and third-neighbor Lennard-Jones parameters are also refined based on quantum-mechanical rotational-energy profiles. The comparison between the UA and AA resolutions is also extended to properties that have not been included as parameterization targets, namely the surface-tension coefficient γ, the isothermal compressibility κ_T, the isobaric thermal-expansion coefficient α_P, the isobaric heat capacity c_P, the static relative dielectric permittivity ϵ, the self-diffusion coefficient D, the shear viscosity η, the hydration free energy ΔG_wat, and the free energy of solvation ΔG_che in cyclohexane. For the target properties ρ_liq and ΔH_vap, the UA and AA resolutions reach very similar levels of accuracy after optimization. For the nine other properties, the AA representation leads to more accurate results in terms of η; comparably accurate results in terms of γ, κ_T, α_P, ϵ, D, and ΔG_che; and less accurate results in terms of c_P and ΔG_wat. This work also represents a first step toward the calibration of a GROMOS-compatible force field at the AA resolution.

Recent progress in general force fields of small molecules

Recent advances in computational hardware and free energy algorithms enable a broader application of molecular simulation of binding interactions between receptors and small-molecule ligands. The underlying molecular mechanics force fields (FFs) for small molecules have also achieved advancements in accuracy, user-friendliness, and speed during the past several years (2018-2020). Besides the expansion of chemical space coverage of ligand-like molecules among major popular classical additive FFs and polarizable FFs, new charge models have been proposed for better accuracy and transferability, new chemical perception of avoiding predefined atom types have been applied, and new automated parameterization toolkits, including machine learning approaches, have been developed for users' convenience.

Diverse Scientific Benchmarks for Implicit Membrane Energy Functions

Energy functions are fundamental to biomolecular modeling. Their success depends on robust physical formalisms, efficient optimization, and high-resolution data for training and validation. Over the past 20 years, progress in each area has advanced soluble protein energy functions. Yet, energy functions for membrane proteins lag behind due to sparse and low-quality data, leading to overfit tools. To overcome this challenge, we assembled a suite of 12 tests on independent data sets varying in size, diversity, and resolution. The tests probe an energy function's ability to capture membrane protein orientation, stability, sequence, and structure. Here, we present the tests and use the franklin2019 energy function to demonstrate them. We then identify areas for energy function improvement and discuss potential future integration with machine-learning-based optimization methods. The tests are available through the Rosetta Benchmark Server (https://benchmark.graylab.jhu.edu/) and GitHub (https://github.com/rfalford12/Implicit-Membrane-Energy-Function-Benchmark).

Evaluation of nine condensed-phase force fields of the GROMOS, CHARMM, OPLS, AMBER, and OpenFF families against experimental cross-solvation free energies

Experimental solvation free energies are nowadays commonly included as target properties in the validation of condensed-phase force fields, sometimes even in their calibration. In a previous article [Kashefolgheta et al., J. Chem. Theory. Comput., 2020, 16, 7556-7580], we showed how the involved comparison between experimental and simulation results could be made more systematic by considering a full matrix of cross-solvation free energies . For a set of N molecules that are all in the liquid state under ambient conditions, such a matrix encompasses N×N entries for considering each of the N molecules either as solute (A) or as solvent (B). In the quoted study, a cross-solvation matrix of 25 × 25 experimental value was introduced, considering 25 small molecules representative for alkanes, chloroalkanes, ethers, ketones, esters, alcohols, amines, and amides. This experimental data was used to compare the relative accuracies of four popular condensed-phase force fields, namely GROMOS-2016H66, OPLS-AA, AMBER-GAFF, and CHARMM-CGenFF. In the present work, the comparison is extended to five additional force fields, namely GROMOS-54A7, GROMOS-ATB, OPLS-LBCC, AMBER-GAFF2, and OpenFF. Considering these nine force fields, the correlation coefficients between experimental values and simulation results range from 0.76 to 0.88, the root-mean-square errors (RMSEs) from 2.9 to 4.8 kJ mol-1, and average errors (AVEEs) from -1.5 to +1.0 kJ mol-1. In terms of RMSEs, GROMOS-2016H66 and OPLS-AA present the best accuracy (2.9 kJ mol-1), followed by OPLS-LBCC, AMBER-GAFF2, AMBER-GAFF, and OpenFF (3.3 to 3.6 kJ mol-1), and then by GROMOS-54A7, CHARM-CGenFF, and GROMOS-ATB (4.0 to 4.8 kJ mol-1). These differences are statistically significant but not very pronounced, and are distributed rather heterogeneously over the set of compounds within the different force fields.

Data-driven analysis of the number of Lennard-Jones types needed in a force field

Force fields used in molecular simulations contain numerical parameters, such as Lennard-Jones (LJ) parameters, which are assigned to the atoms in a molecule based on a classification of their chemical environments. The number of classes, or types, should be no more than needed to maximize agreement with experiment, as parsimony avoids overfitting and simplifies parameter optimization. However, types have historically been crafted based largely on chemical intuition, so current force fields may contain more types than needed. In this study, we seek the minimum number of LJ parameter types needed to represent key properties of organic liquids. We find that highly competitive force field accuracy is obtained with minimalist sets of LJ types; e.g. two H types and one type apiece for C, O, and N atoms. We also find that the fitness surface has multiple minima, which can lead to local trapping of the optimizer.

Molecular engineering of the last-generation CNTs in smart cancer therapy by grafting PEG-PLGA-riboflavin

In this work, the effect of environment and additives on the self-assembly and delivery of doxorubicin (DOX) have been studied. A microfluidic system with better control over molecular interactions and high surface to volume ratio has superior performance in comparison to the bulk system. Moreover, carbon nanotube (CNT) and CNT-doped structures have a high surface area to incorporate the DOX molecules into a polymer and the presence of functional groups can influence the polymer-drug interactions. In this work, the interactions of DOX with both the polymeric complex and the nanotube structure have been investigated. For quantification of the interactions, H-bonding, gyration radius, root-mean-square deviation (RMSD), Gibbs free energy, radial distribution function (RDF), energy, and Solvent Accessible Surface Area (SASA) analyses have been performed. The most stable micelle-DOX interaction is attributed to the presence of BCN in the microfluidic system according to the gyration radius and RMSD. Meanwhile, for DOX-doped CNT interaction the phosphorus-doped CNT in the microfluidic system is more stable. The highest electrostatic interaction can be seen between polymeric micelles and DOX in the presence of BCN. For nanotube-drug interaction, phosphorus-doped carbon nanotubes in the microfluidic system have the largest electrostatic interaction with the DOX. RDF results show that in the microfluidic system, nanotube-DOX affinity is larger than that of nanotube-micelle.

Recent progress on cheminformatics approaches to epigenetic drug discovery

The ability of epigenetic markers to affect genome function has enabled transformative changes in drug discovery, especially in cancer and other emerging therapeutic areas. Concordant with the introduction of the term 'epi-informatics', the size of the epigenetically relevant chemical space has grown substantially and so did the number of applications of cheminformatic methods to epigenetics. Recent progress in epi-informatics has improved our understanding of the structure-epigenetic activity relationships and boosted the development of models predicting novel epigenetic agents. Herein, we review the advances in computational approaches to drug discovery of small molecules with epigenetic modulation profiles, summarize the current chemogenomics data available for epigenetic targets, and provide a perspective on the greater utility of biomedical knowledge mining as a means to advance the epigenetic drug discovery.

Systematic design of biomolecular force fields

Force fields for the study of biomolecules have been developed in a predominantly organic manner by regular updates over half a century. Together with better algorithms and advances in computer technology, force fields have improved to yield more robust predictions. However, there are also indications to suggest that intramolecular energy functions have not become better and that there still is room for improvement. In this review, systematic efforts toward development of novel force fields from scratch are described. This includes an estimate of the complexity of the problem and the prerequisites in the form of data and algorithms. It is suggested that in order to make progress, an effort is needed to standardize reference data and force field validation benchmarks.

Current advances on the development of BET inhibitors: insights from computational methods

Epigenetics was coined almost 70 years ago for the description of heritable phenotype without altering DNA sequences. Research on the field has uncovered significant roles of such mechanisms, that account for the biogenesis of several diseases. Further studies have led the way for drug development which targets epi-enzymes, mainly for cancer treatment. Of the numerous epi-targets involved with histone acetylation, bromodomains have captured the spotlight of drug discovery focused on novel therapies. However, due to high sequence identity, the development of potent and selective inhibitors poses a significant challenge. Herein, we discuss recent computational developments on BET inhibitors and other methods that may be applied for drug discovery in general. As a proof-of-concept, we discuss a virtual screening to identify novel BET inhibitors based on coumarin derivatives. From public data, we identified putative structure-activity relationships of coumarin scaffold and propose R-group modifications for BET selectivity. Results showed that the optimization and design of novel coumarins could be further explored.

Molecular Perception for Visualization and Computation: The Proxima Library

Proxima is a molecular perception library designed with a double purpose: to be used with immersive molecular viewers (thus providing any required feature not supported by third party libraries) and to be integrated in workflow managers thus providing the functionalities needed for the first steps of molecular modeling studies. It thus stands at the boundary between visualization and computation. The purpose of the present article is to provide a general introduction to the first release of Proxima, describe its most significant features, and highlight its performance by means of some case studies. The current version of Proxima is available for evaluation purposes at https://bitbucket.org/sns-smartlab/proxima/src/master/.

Towards Molecular Simulations that are Transparent, Reproducible, Usable By Others, and Extensible (TRUE)

Systems composed of soft matter (e.g., liquids, polymers, foams, gels, colloids, and most biological materials) are ubiquitous in science and engineering, but molecular simulations of such systems pose particular computational challenges, requiring time and/or ensemble-averaged data to be collected over long simulation trajectories for property evaluation. Performing a molecular simulation of a soft matter system involves multiple steps, which have traditionally been performed by researchers in a "bespoke" fashion, resulting in many published soft matter simulations not being reproducible based on the information provided in the publications. To address the issue of reproducibility and to provide tools for computational screening, we have been developing the open-source Molecular Simulation and Design Framework (MoSDeF) software suite. In this paper, we propose a set of principles to create Transparent, Reproducible, Usable by others, and Extensible (TRUE) molecular simulations. MoSDeF facilitates the publication and dissemination of TRUE simulations by automating many of the critical steps in molecular simulation, thus enhancing their reproducibility. We provide several examples of TRUE molecular simulations: All of the steps involved in creating, running and extracting properties from the simulations are distributed on open-source platforms (within MoSDeF and on GitHub), thus meeting the definition of TRUE simulations.

Data-Driven Mapping of Gas-Phase Quantum Calculations to General Force Field Lennard-Jones Parameters

Molecular dynamics simulations are helpful tools for a range of applications, ranging from drug discovery to protein structure determination. The successful use of this technology largely depends on the potential function, or force field, used to determine the potential energy at each configuration of the system. Most force fields encode all of the relevant parameters to be used in distinct atom types, each associated with parameters for all parts of the force field, typically bond stretches, angle bends, torsions, and nonbonded terms accounting for van der Waals and electrostatic interactions. Much attention has been paid to the nonbonded parameters and their derivation, which are important in particular due to their governance of noncovalent interactions, such as protein-ligand binding. Parametrization involves adjusting the nonbonded parameters to minimize the error between simulation results and experimental properties, such as heats of vaporization and densities of neat liquids. In this setting, determining the best set of atom types is far from trivial, and the large number of parameters to be fit for the atom types in a typical force field can make it difficult to approach a true optimum. Here, we utilize a previously described Minimal Basis Iterative Stockholder (MBIS) method to carry out an atoms-in-molecules partitioning of electron densities. Information from these atomic densities is then mapped to Lennard-Jones parameters using a set of mapping parameters much smaller than the typical number of atom types in a force field. This approach is advantageous in two ways: it eliminates atom types by allowing each atom to have unique Lennard-Jones parameters, and it greatly reduces the number of parameters to be optimized. We show that this approach yields results comparable to those obtained with the typed GAFF 1.7 force field, even when trained on a relatively small amount of experimental data.

How to learn from inconsistencies: Integrating molecular simulations with experimental data

Molecular simulations and biophysical experiments can be used to provide independent and complementary insights into the molecular origin of biological processes. A particularly useful strategy is to use molecular simulations as a modeling tool to interpret experimental measurements, and to use experimental data to refine our biophysical models. Thus, explicit integration and synergy between molecular simulations and experiments is fundamental for furthering our understanding of biological processes. This is especially true in the case where discrepancies between measured and simulated observables emerge. In this chapter, we provide an overview of some of the core ideas behind methods that were developed to improve the consistency between experimental information and numerical predictions. We distinguish between situations where experiments are used to refine our understanding and models of specific systems, and situations where experiments are used more generally to refine transferable models. We discuss different philosophies and attempt to unify them in a single framework. Until now, such integration between experiments and simulations have mostly been applied to equilibrium data, and we discuss more recent developments aimed to analyze time-dependent or time-resolved data.

Comparative Exploratory Analysis of Intrinsically Disordered Protein Dynamics Using Machine Learning and Network Analytic Methods

Simulations of intrinsically disordered proteins (IDPs) pose numerous challenges to comparative analysis, prominently including highly dynamic conformational states and a lack of well-defined secondary structure. Machine learning (ML) algorithms are especially effective at discriminating among high-dimensional inputs whose differences are extremely subtle, making them well suited to the study of IDPs. In this work, we apply various ML techniques, including support vector machines (SVM) and clustering, as well as related methods such as principal component analysis (PCA) and protein structure network (PSN) analysis, to the problem of uncovering differences between configurational data from molecular dynamics simulations of two variants of the same IDP. We examine molecular dynamics (MD) trajectories of wild-type amyloid beta (Aβ₁₋₄₀) and its “Arctic” variant (E22G), systems that play a central role in the etiology of Alzheimer's disease. Our analyses demonstrate ways in which ML and related approaches can be used to elucidate subtle differences between these proteins, including transient structure that is poorly captured by conventional metrics.

Deriving Force-Field Parameters from First Principles Using a Polarizable and Higher Order Dispersion Model

In this work we propose a strategy based on quantum mechanical (QM) calculations to parametrize a polarizable force field for use in molecular dynamics (MD) simulations. We investigate the use of multiple atoms-in-molecules (AIM) strategies to partition QM determined molecular electron densities into atomic subregions. The partitioned atomic densities are subsequently used to compute atomic dispersion coefficients from effective exchange-hole-dipole moment (XDM) calculations. In order to derive values for the repulsive van der Waals parameters from first principles, we use a simple volume relation to scale effective atomic radii. Explicit inclusion of higher order dispersion coefficients was tested for a series of alkanes, and we show that combining C6 and C8 attractive terms together with a C11 repulsive potential yields satisfying models when used in combination with our van der Waals parameters and electrostatic and bonded parameters as directly obtained from quantum calculations as well. This result highlights that explicit inclusion of higher order dispersion terms could be viable in simulation, and it suggests that currently available QM analysis methods allow for first-principles parametrization of molecular mechanics models.